Conventional power steering device for a vehicle include an electric power steering device (EPS) provided with a motor as a drive source. Generally, an EPS computes a basic component assist force applied to the steering system, or a basic assist control amount, based on a detected steering torque. In the computation of the basic assist control amount, a dead zone is established in which, when the steering torque is in a predetermined range, the basic assist control amount is set to zero regardless of the value of the steering torque. Thus, in a range including the neutral steering position, excessive assist force is not applied, and the rigid feel of steering is increased.
The surfaces of roads for vehicles are often inclined along the transverse direction in order to improve drainage performance. When a vehicle is traveling forward in a straight line on such an inclined road surface, or on a canted road, the steering wheel needs to be continuously manipulated by small steering angles so that the vehicle is not deflected by the inclination (cant).
However, the steering torque applied in such a case is so small that it is within the dead zone. Therefore, the driver must continue steering without benefitting from the power assist control. This increases the burden on the driver particularly when the vehicle keeps traveling on such a canted road for a long time.
To eliminate such a problem, a method has been proposed in which the condition of cant of the road is detected, and a control component is computed based on the detection result in order to prevent the vehicle from deflecting. The vehicle control apparatus of Patent Document 1 learns the cant of a road based on a plurality of detected state quantities such as the vehicle speed, lateral acceleration, steering state, and driving environment information. The vehicle control apparatus determines the state of cant by means of neural network computation using the results of the learning, thereby computing a control component for reducing the deflection of the vehicle due to the cant.
However, according to the conventional configuration, the computation load is great when accurately determining the state of the cant. This increases the costs. Also, the apparatus has a complicated structure.    Patent Document 1: Japanese Laid-Open Patent Publication No. 2007-22169